Journal of Cheminformatics最新文献

{"title":"Bridging predictive reliability and explainability: a multi-representation deep learning framework for chemical space analysis of immune bioassays","authors":"V. A. Jyothy,&nbsp;Maya L. Pai,&nbsp;E. PA. Sandesh,&nbsp;U. C. A. Jaleel","doi":"10.1186/s13321-026-01206-5","DOIUrl":"10.1186/s13321-026-01206-5","url":null,"abstract":"<div><p>Convergence of chemical and biological space into a unified decision space requires coordinated interactions across the triad- molecular representations, learning algorithms, and explainability. This study establishes a benchmarking framework for Virtual Screening (VS) that integrates this triad to enable consistent, predictive, and interpretable modelling of complex immune-related bioassays. The study employed a wide range of Machine Learning (ML) and Deep Learning (DL) models across three molecular input modalities—descriptors, images, and graphs—while prioritizing predictive reliability and model interpretability. Among classical ML models, Support Vector Machines (SVM) demonstrated the strongest overall performance, while the AttentiveFP architecture outperformed other DL models. Beyond identifying context-specific optimal VS networks, this study derives latent learned concepts that support subsequent hypothesis generation and experimental validation of the interaction between immune targets and ligands.</p><p><b>Scientific contribution</b></p><p>This study establishes a benchmarking framework for virtual screening of immune targets by curating representative bioassay datasets, systematically evaluating multiple molecular representations and learning algorithms, and integrating explainability into the best-performing models. By combining Concept Whitening with the AttentiveFP architecture, we introduce an integrated explainable DL framework that maintains predictive reliability while enabling alignment and separation of domain-specific concepts for immune-target virtual screening. In addition, the study highlights the potential for decision- and logical-level model ensembling, identifies target-specific concept relevance in decision-making, and demonstrates the relationship between data characteristics and model behaviour.</p></div>","PeriodicalId":617,"journal":{"name":"Journal of Cheminformatics","volume":"18 1","pages":""},"PeriodicalIF":7.9,"publicationDate":"2026-05-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1186/s13321-026-01206-5.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147925404","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"化学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"De novo design of anticancer 4-thiazolidinone derivatives: a generative framework shaped by activity cliffs","authors":"Tomasz Szostek,&nbsp;Maciej Wiśniewski,&nbsp;Davide Ballabio,&nbsp;Viviana Consonni,&nbsp;Dariusz Plewczyński,&nbsp;Daniel Szulczyk","doi":"10.1186/s13321-026-01216-3","DOIUrl":"10.1186/s13321-026-01216-3","url":null,"abstract":"<div><p>Activity cliffs (AC) correspond to large potency differences between highly similar compounds and pose a persistent challenge for both predictive modeling and de novo molecular design, particularly in small and underexplored areas of the chemical space. In this study, we introduce an AC-aware generative framework for the de novo design of anticancer thiazolidinone derivatives relevant to non-small cell lung cancer. We identify difficult regions of structure–activity landscape and extract chemical pattern signals hidden in its discontinuities directly from AC pairs using a dedicated algorithm. These signals are incorporated into a fragment-based generative pipeline to guide molecular construction and candidate selection. Compared with a standard QSAR-guided approach, the proposed framework yields a higher fraction of candidates with favorable docking profiles across multiple cancer-relevant targets and produces ligands with more stable binding modes in molecular dynamics simulations. Among the top five candidates selected by a molecular dynamics, multi-metric consensus combining docking affinity, binding stability (ligand and pocket RMSD/RMSF), and energetic criteria, four were generated using the AC-aware CAFE LATE strategy. The two highest-ranked hits exhibited binding energies of − 7.78 and − 7.70 kcal/mol, markedly more favorable than the native reference (− 4.09 kcal/mol), and consistently outperformed native ligands across all evaluated stability metrics. Those results suggest that explicit use of AC information can improve the quality of de novo generated small-molecule candidates in data-limited settings.</p></div>","PeriodicalId":617,"journal":{"name":"Journal of Cheminformatics","volume":"18 1","pages":""},"PeriodicalIF":7.9,"publicationDate":"2026-05-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1186/s13321-026-01216-3.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147872858","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"化学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Prediction of intrinsic solubility for drug-like organic compounds using automated network optimizer (ANO) for physicochemical feature and hyperparameter optimization","authors":"You Kyoung Chung,&nbsp;Seung Jin Lee,&nbsp;Junho Lee,&nbsp;Himchanvit Cho,&nbsp;Sung-Jin Kim,&nbsp;Joonsuk Huh","doi":"10.1186/s13321-026-01220-7","DOIUrl":"10.1186/s13321-026-01220-7","url":null,"abstract":"<div><p>Accurate prediction of aqueous solubility remains a critical challenge in the chemical and pharmaceutical industries, significantly influencing drug development and delivery. This study revisits this well-explored area by leveraging the advanced capabilities of modern computational resources. We apply an automated network optimizer model that integrates dual optimization processes for molecular features and hyperparameters, streamlining the traditionally complex hyperparameter search while providing an efficient interpretation of molecular properties. By employing feature optimization techniques, our deep neural network model demonstrates improvements in both the speed and accuracy of molecular property predictions, achieving an average performance of R² = 0.991. This result outperforms conventional hyperparameter optimization methods such as grid search and random search in predicting the intrinsic solubility of 3,745 compounds across four external experimental datasets. Based on feature importance analysis, we identified key molecular features and structures that significantly influence solubility. Additionally, combining three molecular fingerprints (Morgan, MACCS key, and Avalon) with molecular descriptors enhances model performance, providing a deeper understanding of the relationship between molecular structure and solubility within the physicochemical feature optimization process. These findings underscore the potential of machine learning models to improve predictive modeling of physical properties, apply automated modeling and feature selection to new chemical datasets, and offer explainable insights into the principles driving solubility predictions.</p><p><b>Scientific contributions</b></p><p>This article focuses on recent advancements in the prediction of molecular properties through the application of a Quantitative Structure-Property Relationship (QSPR)-based deep neural network (DNN) model. This model employs a dual optimization approach that integrates both molecular feature selection and hyperparameter tuning. A review of publicly available datasets for drug-sized molecules is presented, highlighting the contributions of automated modeling and feature selection in enhancing the predictive accuracy of physical properties. Additionally, the article addresses the efficacy of these machine learning models in optimizing features, an essential consideration for practical applications in the field.</p></div>","PeriodicalId":617,"journal":{"name":"Journal of Cheminformatics","volume":"18 1","pages":""},"PeriodicalIF":7.9,"publicationDate":"2026-05-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1186/s13321-026-01220-7.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147863348","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"化学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Scaffold-based evaluation metrics for fair comparison of molecular generators","authors":"Valeriia Fil,&nbsp;Remco L. Van Den Broek,&nbsp;Martin Šícho,&nbsp;Ivan Čmelo,&nbsp;M. Isabel Agea,&nbsp;Willem Jespers,&nbsp;Gerard J. P. Van Westen,&nbsp;Daniel Svozil","doi":"10.1186/s13321-026-01213-6","DOIUrl":"10.1186/s13321-026-01213-6","url":null,"abstract":"<div><p>Molecular generators enable the exploration of chemical space to identify novel compounds with desirable properties. However, assessing their performance remains challenging due to the structural diversity and volume of the generated molecules. Commonly used evaluation metrics, focusing on chemical validity and novelty, do not fully align with the primary goal of molecular generation: the discovery of new biologically active compounds. To address this limitation, we introduce scaffold-based metrics that enable a fair comparison by evaluating a generator’s ability to recover biologically relevant scaffolds absent from the input set. We applied the scaffold Recovery Score (RS), SEt scaffold Diversity (SED), and Absolute SEt scaffold Recall (ASER) metrics to compare several molecular generators, including Molpher, DrugEx, REINVENT, and Graph-based genetic algorithm. The proposed scaffold-based metrics provide a realistic framework for evaluating and optimizing molecular generators for their practical use in drug discovery scenarios, particularly in the design of focused virtual chemical libraries. The metrics are available as open-source in a GitHub repository at https://github.com/filvaleriia/scaffold-based-metrics. </p></div>","PeriodicalId":617,"journal":{"name":"Journal of Cheminformatics","volume":"18 1","pages":""},"PeriodicalIF":7.9,"publicationDate":"2026-05-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1186/s13321-026-01213-6.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147855574","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"化学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"GRIPHIN: grids of pharmacophore interaction fields for affinity prediction","authors":"Daniel Rose,&nbsp;Thomas Seidel,&nbsp;Thierry Langer","doi":"10.1186/s13321-026-01203-8","DOIUrl":"10.1186/s13321-026-01203-8","url":null,"abstract":"<div><p>Pharmacophores are widely used to describe protein-ligand interactions. In this work, we propose a hybrid framework for binding affinity prediction that combines pharmacophoric maps of the protein binding site with a graph-based representation of the ligand. Our method achieves performance comparable to state-of-the-art models while offering interpretability through attribution methods, thereby demonstrating the potential of pharmacophoric representations in deep learning.</p><p><b>Scientific contribution</b></p><p>We investigate whether a purely pharmacophoric representation of the protein pocket is sufficient to train a deep learning model for affinity prediction. For this purpose, we devise a hybrid model architecture from simple building blocks for affinity prediction. To enhance interpretability, we apply integrated gradients to attribute predictions to individual pharmacophoric features. Source code and model weights are available at https://github.com/molinfo-vienna/GRIPHIN.</p></div>","PeriodicalId":617,"journal":{"name":"Journal of Cheminformatics","volume":"18 1","pages":""},"PeriodicalIF":7.9,"publicationDate":"2026-05-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1186/s13321-026-01203-8.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147925373","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"化学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"A general-purpose framework for chemical reaction representation with atomic correspondence and flexible condition adaptation","authors":"Kaipeng Zeng,&nbsp;Xianbin Liu,&nbsp;Yu Zhang,&nbsp;Xiaokang Yang,&nbsp;Yaohui Jin,&nbsp;Yanyan Xu","doi":"10.1186/s13321-026-01201-w","DOIUrl":"10.1186/s13321-026-01201-w","url":null,"abstract":"<div><h3>Motivation</h3><p>Organic synthesis is fundamental to the chemical industry, particularly in domains such as pharmaceutical development. While artificial intelligence offers powerful tools for modeling chemical reactions, current approaches are primarily limited to two paradigms: those that rely on hand-crafted, domain-specific features, and those that apply generic deep learning models through simplistic concatenation or aggregation of reaction components. The former often struggles to scale effectively with increasing data volumes, while the latter relies on simple input- or feature-level concatenation to combine different reaction components. Such simplistic aggregation prevents these models from directly capturing the structural transformations between reactants and products, and also makes it difficult to adapt or extend them to datasets that include non-molecular reaction conditions without modifying their model architectures.</p><h3>Results</h3><p>This paper introduces AlignReact, a novel chemical reaction representation learning framework designed for a wide range of organic reaction tasks. Our approach integrates atomic correspondence between reactants and products to discern precise molecular transformations, thereby improving the model’s comprehension of molecular transformation patterns. We incorporate an adapter structure to embed reaction conditions into the representation, enhancing adaptability across varied datasets and tasks. Furthermore, a Reaction-Center-Aware attention mechanism is proposed to enable the model to focus on critical functional groups, yielding more powerful and informative representations. Evaluated across multiple downstream tasks, our model demonstrates superior performance, significantly outperforming existing chemical reaction representation learning architectures on most benchmark datasets.</p><h3>Scientific contribution</h3><p>We introduce a chemical reaction representation learning framework that explicitly integrates atomic correspondence between reactants and products into the network architecture, enabling the model to perceive and model molecular structural transformations during reactions. As an extensible but preliminary feature, our approach also features a flexible, detachable module for integrating reaction conditions. Inspired by conditioning mechanisms from multimodal generation, this adapter module accepts precomputed features of varying dimensions and modalities, laying the groundwork for broader dataset compatibility compared to prior work, and enables fine-grained, context-aware conditioning. Extensive experiments on a range of downstream datasets demonstrate that our framework achieves state-of-the-art performance across several key chemical reaction prediction tasks.</p></div>","PeriodicalId":617,"journal":{"name":"Journal of Cheminformatics","volume":"18 1","pages":""},"PeriodicalIF":7.9,"publicationDate":"2026-05-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1186/s13321-026-01201-w.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147831951","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"化学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Framework for evaluating explainable AI in antimicrobial drug discovery","authors":"Abdulmujeeb T. Onawole,&nbsp;Mark A. T. Blaskovich,&nbsp;Johannes Zuegg","doi":"10.1186/s13321-026-01200-x","DOIUrl":"10.1186/s13321-026-01200-x","url":null,"abstract":"<p>Explainable artificial intelligence (XAI) methods for molecular property prediction lack standardized evaluation criteria, preventing widespread deployment in drug development and hit optimisation, where proper understanding of structure–activity relationship is essential. We developed an evaluation framework for XAI using fragment-based explainability tests to compare XAI with different molecular representation and challenge the different XAI approaches for proper explanation of activity cliffs. The evaluation methods include essential scaffold recognition, scaffold sensitivity and substructure specificity for explaining activity cliff, and technical evaluation on model robustness and consistency. Using a curated dataset of antibiotic molecules we established three XAI models with fundamentally different molecular representation: Random Forest on chemical features using SHAP, CNN on sequence-based SMILES using token occlusion, and RGCN on molecular graphs with substructure masking. Together with detailed case study, we evaluated the explainability behaviours and quality of the different XAI approaches and highlighted their limitations. While all XAI approaches displayed good predictive and scaffold recognition capabilities, and comparable robustness and consistency, they displayed quite different explainability behaviour for activity cliffs, revealing their different utility for medicinal chemistry.</p><p>A.T.O. performed the study, A.T.O and J.Z. contributed to the concept of the study and wrote the original manuscript. All authors reviewed the manuscript.</p>","PeriodicalId":617,"journal":{"name":"Journal of Cheminformatics","volume":"18 1","pages":""},"PeriodicalIF":7.9,"publicationDate":"2026-05-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1186/s13321-026-01200-x.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147831991","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"化学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Improving protein-ligand complex generation with force field guidance","authors":"Helen Lai,&nbsp;Tingyu Wang,&nbsp;Hassan Sirelkhatim,&nbsp;Joe Eaton,&nbsp;Howard Huang,&nbsp;Brad Rees,&nbsp;Ola Engkvist,&nbsp;Jon Paul Janet,&nbsp;Xiaoyun Wang,&nbsp;Alessandro Tibo","doi":"10.1186/s13321-026-01198-2","DOIUrl":"10.1186/s13321-026-01198-2","url":null,"abstract":"<p>Generative models based on diffusion and flow matching have recently been applied to structure-based drug design, but their outputs often include unrealistic protein–ligand interactions that do not obey the laws of physics. We present an energy guidance framework that incorporates a molecular mechanics force field (MMFF94) directly into the sampling process. The method steers molecular generation toward more physically plausible and energetically stable conformations without retraining the underlying model. We evaluate this approach using two state-of-the-art architectures, SemlaFlow, a flow matching model and EDM, a diffusion model, on the PDBBind dataset. Across both models, energy guidance improves enthalpic interaction energy, improves strain energy by up to 75<span>(%)</span>, and generates over 1000 ligands with better docking scores than native ligands. These results demonstrate that lightweight, physics-based guidance can significantly enhance generative drug design while preserving chemical validity and diversity.</p><p>We introduce a novel, <i>training-free force field guidance</i> framework that steers ligand generation using empirical molecular mechanics (e.g., MMFF94) during diffusion or flow-based sampling–without modifying or retraining the base generative model (e.g., EDM or Semflaflow by [24]). Our method operates as a plug-in during inference time, leveraging energy feedback to generate poses with lower strain and having better predicted interactions with the protein structure.</p><p>Our main contributions are as follows:</p>","PeriodicalId":617,"journal":{"name":"Journal of Cheminformatics","volume":"18 1","pages":""},"PeriodicalIF":5.7,"publicationDate":"2026-05-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1186/s13321-026-01198-2.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147809073","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"化学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Defining peptides in ChEBI","authors":"Simon Flügel,&nbsp;Till Mossakowski,&nbsp;Fabian Neuhaus,&nbsp;Erik Pfanenstiel,&nbsp;Martin Glauer,&nbsp;Edgar Haak,&nbsp;Adnan Malik,&nbsp;Noel M. O’Boyle","doi":"10.1186/s13321-026-01196-4","DOIUrl":"10.1186/s13321-026-01196-4","url":null,"abstract":"<div><p>Modern biochemistry is producing vast amounts of chemical knowledge. Ontologies, such as the Chemical Entities of Biological Interest (ChEBI) ontology, can help organising this knowledge. With manual classification alone however, ontologies cannot keep up with the growth of their domain. In this work, we propose a novel taxonomy of 67 classes related to peptides, a large branch in ChEBI with nearly 15,000 compounds. The existing natural language definitions in ChEBI have been expanded and specified more precisely. These natural language definitions are accompanied by a logical axiomatisation in monadic second-order logic (MSOL). To use the axiomatisation for automated classification, a methodology has been developed that translates monadic second-order definitions first into partial first-order definitions and finally into an algorithmic classification. This connects three aspects important to ontological definitions: They reflect the opinions of experts, they are unambiguous, and they can be checked automatically. In our evaluation, we compare the results of our classification to the current taxonomy of ChEBI . This reveals potential inconsistencies in ChEBI as well as areas that might benefit from automated extensions. We also evaluate our natural-language definitions in an expert survey.</p><p><b>Scientific contribution:</b> This work provides precise natural-language definitions of 14 current ChEBI classes as well as 53 new peptide-related classes. These definitions are formalised in MSOL and come with an efficient implementation that allows for large-scale molecule classification, including a full classification of ChEBI and PubChem.</p></div>","PeriodicalId":617,"journal":{"name":"Journal of Cheminformatics","volume":"18 1","pages":""},"PeriodicalIF":5.7,"publicationDate":"2026-05-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1186/s13321-026-01196-4.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147808910","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"化学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"AdaptMol: domain adaptation for molecular image recognition with limited supervision","authors":"Feng Hu,&nbsp;Estrid He,&nbsp;Karin Verspoor","doi":"10.1186/s13321-026-01209-2","DOIUrl":"10.1186/s13321-026-01209-2","url":null,"abstract":"<div><p>Optical Chemical Structure Recognition (OCSR) aims to convert two-dimensional molecular images into machine-readable formats such as SMILES strings. Deep learning has substantially improved OCSR performance, yet most methods rely on synthetic training data and struggle to generalize to real-world inputs, especially hand-drawn diagrams, where stroke width, geometry, and drawing conventions vary widely across individuals. In this work, we propose an image-to-graph model AdaptMol that enables effective transfer from synthetic to real-world data without requiring manual graph annotations in the target domains. AdaptMol is an integrated pipeline that starts with training a base model on synthetic data, and then refines model representations through unsupervised domain adaptation and self-training. Our key insight is that bond features are domain-invariant in nature; they encode structural relationships between atoms that are independent of visual variations across domains. Thus, during domain adaptation, we align bond-level feature distributions via class-conditional Maximum Mean Discrepancy (MMD) to enforce cross-domain consistency. We also design a comprehensive data augmentation strategy to enhance the robustness of the base model, facilitating stable self-training on unlabeled target samples. On hand-drawn molecular images, our model achieves 82.6% accuracy and outperforms the best prior method by 10.7 points, while maintaining competitive performance across four benchmarks comprising molecular images from scientific literature and patent documents.</p><p><b>Scientific contribution</b></p><p>We propose AdaptMol, an image-to-graph model that predicts molecular structures as graphs of atoms and bonds, achieving effective transfer from synthetic to hand-drawn molecular images without requiring target domain graph annotations. We combine class-conditional Maximum Mean Discrepancy to align bond features across domains with comprehensive data augmentation to increase training data variation, jointly improving base model accuracy sufficiently for self-training and addressing the critical failure mode of prior approaches that begin with insufficient accuracy. We further introduce a dual position representation that supervises atom positions through both discrete coordinate tokens and continuous spatial heatmaps to reduce false positives in atom localization.</p></div>","PeriodicalId":617,"journal":{"name":"Journal of Cheminformatics","volume":"18 1","pages":""},"PeriodicalIF":7.9,"publicationDate":"2026-05-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1186/s13321-026-01209-2.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147808840","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"化学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}